Lexical Access via Phoneme to Grapheme Conversion

Lexical Access via Phoneme to Grapheme conversion 1Michael Fridkin, 1David L. Dowe and 2Simon Musgrave 1Clayton School of I.T 2School of Languages, Cultures and Linguistics Monash University VIC 3800, Australia mfri7@student.monash.edu.au david.dowe@infotech.monash.edu.au simon.musgrave@arts.monash.edu.au Abstract The Lexical Access (LA) problem in Com- puter Science aims to match a phoneme se- quence produced by the user to a correctly spelled word in a lexicon, with minimal hu- man intervention and in a short amount of time. Lexical Access is useful in the case where the user knows the spoken form of a word but cannot guess its written form or where the users best guess is inappropri- ate for look-up in a standard dictionary or by traditional computer spellcheckers. Pre- vious approaches to this problem have at- tempted to match user-generated phoneme sequences to phoneme sequences in dic- tionary entries and then output the stored spelling for the sequences. Our approach includes a phoneme-to-grapheme conver- sion step followed by spelling-to-spelling alignment. This ensures that more distor- tion to the data can be handled as well as reducing error rates and also allows us to combine the Lexical Access problem with a related task in Natural Language Pro- cessing, that of spelling Out of Vocabulary (OOV) items. We call this combination the Hybrid Lexical Access (HLA) Prob- lem. The performance of our system on this task is assessed by the percentage of correct word matches in the output after edit dis- tance is performed to find the closest match to a stored item. 1 Introduction In languages with irregular spelling systems, such as English, most speakers have trouble spelling at least some words. To aid spelling there are orthographic dictionaries (which index words by their sequence of letters) and phonemic dictio- naries (which index words by their sequence of phonemes). We will be looking at how, given a phoneme sequence, phonemic dictionaries can be used by computer systems to look up single words and if the word is not in the lexicon then how computer systems can best give a suggestion to its true spelling. In orthographic dictionaries, finding the cor- rect word depended on being able to spell the first few letters correctly. Spellcheckers have im- proved on this method but still have two draw- backs. Firstly, they require a candidate spelling before they can begin matching and this can prove difficult in some situations; for example, an un- stressed vowel sound (normally schwa in English) can be spelled with many different graphemes. Secondly, spell-checkers have difficultly match- ing multiple graphemes which can correspond to the same phoneme. For example, in the word Aachen, the candidates Acen, Aken and Acken would not map to the correct spelling using stan- dard spellcheckers. Few speakers of English would know that /k/ is spelled as ch in some cir- cumstances such as the one in the example. Com- puter systems can supplement the user’s linguistic knowledge in these situations. In Computer Science the Lexical Access Prob- lem involves matching a sequence of phonemes to a …

Content truncated to 500 words